fix(opencode): respect configured model input limits

[danielwolfman]

Issue for this PR

Closes #16298
Refs #16308

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

It keeps provider.models.<model>.limit.input when building provider config overrides and allows config files to override just limit.input without restating context and output.

That fixes the long-context OpenAI case where users want to raise the client-side compaction threshold for models like gpt-5.4.

How did you verify your code works?

• Added provider tests that verify custom models keep configured limit.input
• Added a regression test that verifies an openai model override can set only limit.input
• Ran bun test test/provider/provider.test.ts
• Ran bun typecheck in packages/opencode

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Found a related PR:

PR #16306: fix(opencode): honor model limit.input overrides

• fix(opencode): honor model limit.input overrides #16306

This PR appears to address the same issue - honoring/respecting model limit.input configuration overrides. It's likely a closely related or duplicate attempt at fixing the same problem described in the current PR #18070.